KR20020095758A - Adaptive motion predict method - Google Patents

Abstract

PURPOSE: An adaptive motion estimation method is provided to adaptively determine a search area according to the motion characteristics of images, thereby reducing the computation amount for the motion estimation and improving the entire codec performance. CONSTITUTION: An adaptive motion estimation method includes the steps of obtaining an average value of SAD values of previous k-th frame from a current image, comparing an average value for SAD values from (k+1)th to 2k-th frames with the average value of the SAD values of the previous k-th frame from the current frame, and reducing the motion estimation area if that average value from the (k+1)th to 2k-th frames is smaller and increasing the area in the other case in a motion estimation area determining part(111).

Description

Adaptive motion estimation method {ADAPTIVE MOTION PREDICT METHOD}

The present invention relates to an image processing system, and more particularly, to an adaptive motion estimation method in video encoding.

In addition to increasing interest in multimedia application services through a high-speed Internet network, the need for video service using an Asymmetric Digital Subscriber Line (ADSL) network is increasing.

The ADSL network is characterized by being able to use high-speed data communication and regular telephone at the same time while using the current telephone line.

Therefore, high speed video codec implementation is essential for video service through ADSL network.

The video codec can be implemented with ITU-T H.26x (H.263, H.263 +, H.263 ++, H.26L), MPEG2, etc., which is the standard video compression standard for MPEG4 and video telephony systems. All of these techniques use motion estimation in order to take advantage of the temporal correlation of the image, which affects the overall codec performance in terms of complexity, that is, the number of frames that can be encoded per second.

Most video encoders currently standardized or standardized perform frame coding, motion estimation / compensation, and entropy coding, with the first frame (I-frame) being Discrete Cosine Transform (DCT), quantization ( Quantization (VLC) and Variable Length Coding (VLC) are performed, and the remaining frames are motion estimated and compensated through P-frame encoding.

In general, the P-frame encoding process consists of a sequence of motion estimation in units of integer pixels, mode determination for each macro block, and motion estimation in units of half pixels, which will be described below.

First, motion estimation is performed on an integer pixel unit for each macro block, and then, whether the macro block is an INTRA mode or an INTER mode is determined.

Accordingly, when determined in the INTRA mode, the macroblock is encoded in the same manner as I-frame encoding, in which case motion estimation and compensation are not necessary.

On the contrary, when the INTER mode is determined, motion compensation is performed on the macroblock to obtain a motion vector, motion compensation is performed, and the motion compensation error value is encoded in the same manner as in the case of the INTRA mode.

If the motion vector is not obtained or motion estimation and compensation fail, the texture information of the corresponding macroblock is transmitted to the decoder.

Accordingly, the decoder obtains a reconstructed image by adding a prediction block by a motion vector and an encoded error block.

Meanwhile, the motion estimation process includes calculating a sum of the absolute difference (SAD) for estimating a similar degree between the current image and the previous image in a predetermined search area, and the calculation equation is as follows.

Here, SAD (x, y) represents a cumulative prediction error in the search region x, y, and + -15 represents a search region used for motion estimation.

Therefore, the position (x, y) having the smallest SAD value is selected as the final motion vector.

This method applies the same search range for all successive frames and the same for macroblocks.

However, MPEG-4 and ITU-T, H.26x, MPEG2, which are video compression standards for video telephony systems, which are currently being standardized, are commonly fixed when using motion estimation techniques to take advantage of the temporal correlation of video. In the conventional search area, the conventional method has a disadvantage in that a large search area for a small motion image does not properly reflect the motion characteristics of the image and also requires an unnecessary calculation amount in terms of complexity.

Accordingly, in order to implement a high speed image encoder, a technique for adaptively determining a search region according to a motion characteristic of an image is required.

Accordingly, the present invention is to determine the search area by using the cumulative SAD information that is known to reflect the complexity of the image for each frame in order to improve the disadvantages of the conventional and adaptively adjust the size of the search area according to the motion characteristics of the image The purpose of the present invention is to provide an adaptive motion estimation method, which is designed to improve the performance of the overall encoder by reducing the amount of motion estimation that has a significant effect on the number of frames that can be encoded per second.

1 is a block diagram of a motion estimation apparatus for an embodiment of the present invention.

2 is an exemplary view showing a reference image of an adaptive motion estimation region in an embodiment of the present invention.

Explanation of symbols on the main parts of the drawings

101: subtractor 102: switch

103: DCT converter 104: quantizer

105: entropy coding unit 106: inverse quantizer

107: IDCT conversion unit 108: the adder

109: video frame 110: motion estimation unit

111: motion estimation region determiner

In order to achieve the above object, the present invention provides a method of estimating motion by comparing a current image and a previous image, the method comprising: calculating an average value of SAD values up to a previous k-th frame based on the current image; Obtaining a mean value of the SAD values from the th frame to the 2k interval frame, and comparing the average value of the SAD value from the previous kth frame to the k-th frame, and reducing the motion estimation region if the comparison result is small and increasing the motion estimation region if the comparison result is small. Characterized in performing the steps.

Hereinafter, the present invention will be described in detail with reference to the drawings.

1 is a block diagram of an image encoding apparatus for an embodiment of the present invention, as shown in FIG. 1, a subtractor 101, a switch 102, a DCT converter 103, a quantizer 104, and an entropy coding unit ( 105), an inverse quantizer 106, an IDCT converter 107, an adder 108, an image frame 109, a motion estimator 109, and a motion estimation region determiner 111.

Referring to the operation of the embodiment of the present invention configured as described above are as follows.

When the original image is temporarily stored in the frame memory (not shown) and then outputted in units of frames, the subtractor 101 outputs a difference signal between the memory image and the reconstructed image, and the switch 102 selects one of the memory image or the difference signal. It is input to the DCT converter 103.

The DCT transformed result (DCT coefficient) of the DCT converter 103 is quantized by the quantizer 104 and then input to the entropy coding unit 105 and the dequantizer 106.

The inverse quantizer 106 inversely quantizes the quantized signal to obtain an inverse DCT image through the IDCT converter 107 to reconstruct the currently input image.

The reconstructed image is added to the motion-compensated signal by the adder 108, stored in the image memory 109, and then input to the motion estimation unit 110.

The motion estimator 110 predicts a motion of an image to be encoded next, which is output from the image memory 109, and predicts a current image and a motion in a search area determined by the motion estimation region determiner 111. Comparing the desired image to obtain a motion vector (MV), and outputs to the entropy coding unit 105, and outputs the motion-compensated image according to the motion vector (MV) value to the subtractor 101 and the adder 108 Done.

Therefore, by repeatedly performing the above operation, the entropy coding unit 105 codes the final quantized controlled image information and then transmits the final image information to the decoder side along with the motion vector (MV).

Meanwhile, the motion estimator 110 stores the SAD value of each macro block represented by the following equation in the motion estimation process to determine the adaptive motion search region.

Where N is the number of macroblocks per frame, Is the SAD value of each macroblock, Is a value determined at the time of motion estimation of the previous frame and represents the total SAD value accumulated in each frame.

Accordingly, the motion estimation region determiner 111 may determine the previous frames stored in the motion estimation unit 110 to determine the search region of the current frame. The value is used.

The reason for this is that the complexity of the motion feature estimation technique by scene change detection using the difference between the pixels of the previous frame is increased to estimate the motion characteristic of the current frame.

Accordingly, the adaptive search region determination method according to the present invention will be described with reference to FIG.

In Figure 2 Displays the previous video based on the current video, Indicates a previous image than the previous image.

In the exemplary embodiment of the present invention, as shown in FIG. 2, the average value of the SAD values up to the previous kth frame is calculated based on the current image, and the average value of the SAD values up to the second 2k frame is obtained. The average value of is compared.

Accordingly, the motion estimation region determiner 111 searches if the average value of the SAD values up to the 2kth frame is smaller than the average value of the SAD values up to the kth frame based on the current frame image based on the kth frame image. Reducing the area, on the contrary, increases the search area if it is large.

This is because a small SAD value indicates a small section of motion, and a large SAD value indicates a large section of motion.

Meanwhile, the motion estimation region determiner 111 arbitrarily sets a predetermined section k, and then compares the SAD section average value with the SAD average value from the next (k + 1) to the 2k section to determine the search region. Assuming that the size of the region is 'N', the adaptive search region N is determined by the following equation.

here, Is a constant determined experimentally so that a given search range does not exceed the maximum search range, and ROUND represents a rounding operation.

By the way, the interval k for obtaining the average value of the SAD value for motion estimation can be arbitrarily determined, but the smaller the setting interval, the more accurate image motion can be estimated.

In conclusion, the present invention accurately estimates the motion by increasing the search area for frames with large movements, but reduces the computational amount and reduces performance while reducing the search area for frames with small movements.

As described in detail above, the present invention has the effect of improving the performance of the entire codec by adaptively determining the search area according to the motion characteristics of the image to reduce the amount of computation for motion estimation.

Therefore, the present invention can be applied to the implementation of a video phone for a high speed network faster than before.

Claims (4)

A first step of obtaining and comparing a similarity average value of the previous kth frame with a similarity average value of the (k + 1) th to 2kth frames based on the current image; and determining a motion estimation region based on the comparison result And a second step of estimating motion by applying the motion estimation region determined above to the current image and the previous image. The method of claim 1, wherein the similarity mean value of the first step is the SAD value of the entire frame expressed as Adaptive motion estimation method characterized in that the average value of. Where N is the number of macroblocks per frame, Is the SAD value of each macroblock, Is a value determined at the time of motion estimation of the previous frame and represents the total SAD value accumulated in each frame. The method of claim 1, wherein the second step reduces the motion estimation region when the average value of the SAD values from the (k + 1) th to the 2k th frames is smaller than the average value of the SAD values of the previous k th frame based on the current image. On the contrary, the adaptive motion estimation method of claim 1, wherein the motion estimation area is increased in large cases. The method of claim 1 or 3, wherein the motion estimation region (N) is obtained by the following equation. here, Is a constant determined experimentally so that a given search range does not exceed the maximum search range, and ROUND represents a rounding operation.